LTX /v1/extend limits, and chaining clips with Sume frame_images

LTX's POST /v1/extend continues an existing video, but only at 1080p, adding 2 to 20 seconds, with at least 73 input frames and a 505-frame total. Sume does not expose an extend call on its Omni video id. To get a longer video, generate a next clip from the last frame of the previous one with frame_images, then join the clips with Timeline.

LTX's limits are from its post How to make longer AI videos, read 2026-10-01. Sume's side is from Video generation and Timeline 1.0.

What are the /v1/extend limits?

Does Sume have an extend call?

Not on the Omni model. The Video Router docs say previous_interaction_id / extend is not exposed for gemini-omni-flash-1.1, because it is not on the provider's published input schema. The docs read here do not describe an extend call on other rows, so the supported route is chaining.

How do I chain clips on Sume?

Pull the last frame of clip one (see get the last frame to chain clips), then send it as a frame_images entry with frame_type: "first_frame" on the next request. The docs note that if both frame_images and input_references are provided, frame_images takes precedence and the request is treated as image-to-video. Check supported_frame_images in the catalog first, since not every model accepts a frame.

Chaining restarts generation from a still, so motion and audio are not carried over the way an extend carries them. Expect to review each seam.

How do I join the clips?

Timeline 1.0 assembles one audio spine plus ordered video slots into a long-form MP4, per its docs. Import the generated files as Sume-hosted media first, then list them in video[]. The longer walkthrough is extend an AI video past 30 seconds; the Veo equivalent is in Veo 3.1 extend vs chaining.